Informative cluster sizes for subcluster-level covariates and weighted generalized estimating equations.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical modeling

Background:

• Existing methods like cluster-weighted generalized estimating equations (CWGEE) adjust for informative cluster sizes but may fail with within-cluster covariate variations.
• Inverse probability of treatment weighting addresses informative treatment propensity but not informative cluster size.

Purpose of the Study:

• To develop novel weighted generalized estimating equation (GEE) estimators that simultaneously account for informative cluster sizes and within-cluster exposure distributions.
• To evaluate the performance of these new estimators against existing methods in the presence of complex informativeness.

Main Methods:

• Proposed several weighted GEE estimators with weights incorporating cluster size and the distribution of a binary exposure within clusters.
• Conducted simulation studies to compare the performance of the new estimators against GEE, CWGEE, and inverse probability of treatment-weighted GEE.
• Applied the method to analyze covariate effects on dental caries risk in a study of young children.

Main Results:

• The proposed weighted GEE estimators demonstrated superior performance compared to existing methods in simulation studies.
• The new estimators effectively addressed bias arising from both informative cluster sizes and informative within-cluster exposure distributions.
• The method was successfully applied to a real-world example concerning dental caries risk.

Conclusions:

• The developed weighted GEE estimators offer a robust approach for analyzing clustered data with complex informativeness.
• These methods provide improved bias adjustment, leading to more reliable effect estimates in epidemiological and biostatistical research.
• The proposed approach is valuable for studies involving binary exposures and informative cluster characteristics, such as the dental caries example.